Association of oral bisphosphonates with cardioembolic ischemic stroke: a nested case-control study

Background: Bisphosphonates have been reported to increase the risk of atrial fibrillation. Therefore, it is conceivable that they may increase the risk of cardioembolic ischemic stroke (IS). However, most epidemiological studies carried out thus far have not shown an increased risk of IS, though none separated by the main pathophysiologic IS subtype (cardioembolic and non-cardioembolic) which may be crucial. In this study, we tested the hypothesis that the use of oral bisphosphonates increases specifically the risk of cardioembolic IS, and explored the effect of treatment duration, as well as the potential interaction between oral bisphosphonates and calcium supplements and anticoagulants. Methods: We performed a case-control study nested in a cohort of patients aged 40–99 years, using the Spanish primary healthcare database BIFAP, over the period 2002-2015. Incident cases of IS were identified and classified as cardioembolic or non-cardioembolic. Five controls per case were randomly selected, matched for age, sex, and index date (first recording of IS) using an incidence-density sampling. The association of IS (overall and by subtype) with the use of oral bisphosphonates within the last year before index date was assessed by computing the adjusted odds ratios (AOR) and their 95% CI using a conditional logistic regression. Only initiators of oral bisphosphonates were considered. Results: A total of 13,781 incident cases of IS and 65,909 controls were included. The mean age was 74.5 (SD ± 12.4) years and 51.6% were male. Among cases, 3.15% were current users of oral bisphosphonates, while among controls they were 2.62%, yielding an AOR of 1.15 (95% CI:1.01–1.30). Of all cases, 4,568 (33.1%) were classified as cardioembolic IS (matched with 21,697 controls) and 9,213 (66.9%) as non-cardioembolic IS (matched with 44,212 controls) yielding an AOR of 1.35 (95% CI:1.10–1.66) and 1.03 (95% CI: 0.88–1.21), respectively. The association with cardioembolic IS was clearly duration-dependent (AOR≤1 year = 1.10; 95% CI:0.82–1.49; AOR>1–3 years = 1.41; 95% CI:1.01–1.97; AOR>3 years = 1.81; 95% CI:1.25–2.62; p for trend = 0.001) and completely blunted by anticoagulants, even in long-term users (AOR>1 year = 0.59; 0.30–1.16). An interaction between oral bisphosphonates and calcium supplements was suggested. Conclusion: The use of oral bisphosphonates increases specifically the odds of cardioembolic IS, in a duration-dependent manner, while leaves materially unaffected the odds of non-cardioembolic IS.

Supplementary Methods

Data source
BIFAP (Base de datos para la Investigación Farmacoepidemiológica en el Ámbito Público) is a primary care database managed by the Spanish Agency of Medicines and Medical Devices (AEMPS) that contains pseudonymized information on clinical diagnoses, medical comments in free text, laboratory tests, vaccinations, complementary explorations and prescriptions, prospectively recorded by primary care physicians (PCPs) in the National Health System. Clinical events are recorded using the International Classification of Primary Care, version 2 (ICPC-2) or the International Classification of Diseases, version 9, Clinical Modification (ICD-9-CM), depending on the region. All prescriptions written by the PCPs are recorded including product name, quantity, dosing regimens, indication and date of prescription. The BIFAP population is representative of the population receiving healthcare in Spain (Maciá-Martínez et al., 2020). The information in the database is enriched with free text annotations of the PCPs. BIFAP has been extensively validated through multiple pharmacoepidemiological studies in different areas, including cardiovascular (Rodríguez-martín et a., 2020) To conduct this study, the 2016 version was used, which contains data from 7.6 million patients, with an average of 5.1 years of follow-up (38.8 million person-years), from nine different Spanish regions. The study period encompasses 14 years, from January 1, 2002 to December 31, 2015.

Stroke case validation procedures
BIFAP contains information on clinical diagnoses that PCPs record as part of their routine clinical practice. In order to adequately perform pharmacoepidemiological studies in BIFAP, the event of interest recorded in the data source is validated.
In BIFAP database, predefined case-finding algorithms (CFA) for clinical events are available based on proper code selection within ICPC-BIFAP or ICD-9 codes. The methodology to normalize the diagnosis information in BIFAP is detailed elsewhere24.
The validation procedures of the predefined stroke CFA included the following steps: 1. Definition of the study population and identification of potential stroke cases retrieved by the stroke CFA. 2. Manual review of the patient's electronic healthcare records -including clinical notes-, in a random sample of 1000 potential cases, to confirm or rule out the diagnosis according prespecified criteria. This review also included the validity of the date of the event. 3. Estimation of the Positive Predictive Value (PPV) of the predefined stroke CFA to identify valid incident ischemic strokes. 4. Stroke CFA refinement strategies: Using natural language processing (NLP) techniques to increase the PPV in those subsets with PPV lower than 80%.

Validation of the predefined stroke CFA in BIFAP
Predefined CFA for ischemic stroke, hemorrhagic stroke or unspecified stroke are available in BIFAP database. After excluding hemorrhagic strokes, we identified through the CFA 24094 ischemic and 15868 unspecified strokes identified, totaling 39962 cases.
Of them, a sample of 1000 cases was randomly selected for validation purposes. The electronic healthcare records (EHR) -including clinical notes-of the sample were manually reviewed, blinded to drug exposure, independently by two investigators (SRM and DBH) looking for additional information to confirm that it was a true case of incident stroke. Discrepancies were settled by the entire research group. According to this, patients were classified according to pre-specified criteria as: • Valid cases: when additional evidence was found in the EHRs that confirmed the diagnosis; • Cases with insufficient information: when there was no additional information in the EHRs to support or rule-out the stroke diagnosis; • Non-case: when the available information allowed to rule out the diagnosis of incident stroke.
The review of the random sample of 1000 potential stroke cases showed the following results: 641 valid cases; 168 insufficient information and 191 non-cases.

Refinement procedures to increase the PPV of the predefined ischemic stroke CFA.
Natural language processing techniques were implemented, in order to identify: a) Semantic patterns in clinical notes with high likelihood to be in the clinical notes of the valid stroke cases. b) Semantic patterns in clinical notes with high likelihood to be in the clinical notes of the noncases and low likelihood to be in the clinical notes of valid cases.
Then, semantic patterns identified in (a) were added to the pre-defined stroke CFA as additional criteria for stroke case selection and semantic patterns identified in (b) were used to consider them as non-cases.
Using this refinement, the resulting PPV in the random sample increased to 87,1% and consequently, the refined incident stroke CFA was applied to all potential stroke cases. A total number of 14374 cases was included in the case-control study.

Identification of the main pathophysiological subtype among ischemic stroke cases
In this step, the free text associated with the diagnosis was examined. The validation strategy focused on identifying and exclude from the study stroke cases of probable cardioembolic origin, as well as other rare causes (e.g. vasculitis, vascular dissection, drug abuse, etc,).
The following main criteria were applied to identify cardioembolic stroke cases:  Cardioembolic text criterion: the word "cardioembolic "or other related terms was sought in text comments associated with the stroke diagnosis.  Atrial fibrillation criterion: patients with a diagnosis of atrial fibrillation prior to the stroke event or within three months of stroke diagnosis were identified.  Oral anticoagulant use criterion: patients on oral anticoagulant therapy at the time of stroke diagnosis (or up to three months prior to diagnosis) or starting treatment in a three-month window since stroke diagnosis.
The following additional criteria were used when at least one of the above was met:  Antiarrhythmic use criterion: patients on treatment with antiarrhythmic drugs (class IC and III) at the time of stroke diagnosis (or up to three months prior to diagnosis) or starting treatment in a three-month window since stroke diagnosis. In order to consider a case as cardioembolic stroke, it was required to have some other of the aforementioned criteria in addition to the use of antiarrhythmic drugs.  Mitral valve prosthesis/stenosis criterion: patients with a record of mechanical valve prosthesis or mitral stenosis prior to stroke diagnosis or up to three months after the event. In order to consider a case as cardioembolic, it was required to have some other of the aforementioned criteria in addition to the record of mitral valve prosthesis or stenosis.
Regardless the criteria listed above, all patients who had text strings in their commentaries associated with the stroke diagnosis compatible with the words atherothrombotic, thrombogenic, or lacunar (including related terms) were classified as non-cardioembolic stroke.
Patients with a free-text describing a vascular dissection, cocaine abuse and vasculitis as the probable cause of the stroke were identified in a distinct category and excluded from most analyses.
Ischemic stroke cases classified as probable cardioembolic origin according to the criteria applied.  The arrows connecting bisphosphonates with Atrial cardiopathy and AF (clinical and subclinical) and AF with CIS represent the postulated causal pathway. The possibility that AF can be treated with OACs, a strong protective factor of CIS, is also reflected in the causal graph. The square surrounding L, denotes that the effect of bisphosphonates on CIS is conditioned (or adjusted) on such variables. -Angina pectoris: recorded as such, and/or when patients were using nitrates.

Number of cases
-Diabetes: recorded as such, and/or when patients were using glucose-lowering drugs.
-Dyslipidemia: recorded as such, and/or when patients were using lipid-lowering drugs.

Supplementary Figure 3.
Oral bisphosphonates and risk of non-cardioembolic ischemic stroke by age groups, sex, background vascular risk and CHA2DS2-VASC score.
Definitions of different categories of vascular risk: (1) established vascular disease: those with a history of ischemic heart disease (AMI or angina pectoris), heart failure, transient ischemic attack, peripheral arterial disease or diabetes; (2) one or more vascular risk factors: those with a history of hypertension, dyslipidemia, chronic renal failure, current smoking, or body mass index >30 kg/m2 (and none of the conditions mentioned in the first point); (3) no known vascular risk factor or disease: the remainder.

Supplementary Tables
Supplementary * Adjusted only for matching factors (age, sex, and calendar year). † Recorded as such, and/or when patients were using nitrates. § Recorded as such, and/or when patients were using glucose-lowering drugs. ⁋ Recorded as such, and/or when patients were using lipid-lowering drugs. ‡ Including tibolone. heart disease (including history of acute myocardial infarction or angina pectoris -recorded as such and/or use of nitrates), thromboembolism, heart failure, peripheral artery disease, hypertension, diabetes (recorded as such, and/or use of glucose-lowering drugs), dyslipidemia (registered as such, and/or use of lipid-lowering drugs), hyperuricemia (asymptomatic and gout), chronic obstructive pulmonary disease, rheumatoid arthritis, and chronic renal failure, and use within the last 30 days of antiplatelet agents, beta-blockers, alpha blockers, angiotensin-converting enzyme inhibitors, angiotensin II receptor antagonists, calcium antagonists, diuretics, paracetamol, metamizole, non-steroidal anti-inflammatory drugs, corticosteroids, opioids, calcium with/without vitamin D supplements, hormonal replacement therapy, estrogen receptor modulators, strontium ranelate, calcitonin, denosumab, teriparatide, proton pump inhibitors and H2-receptor antagonists.